Jeff Pelletier

A certain direction in cognitive science has been to try to “ground” public language statements in some species of mental representation. A central tenet of this trend is that communication – that is, public language – succeeds (when it does) because the elements of this public language are in some way correlated with mental items of both the speaker and the audience so that the mental state evoked in the audience by the use of that piece of public language is the one that the speaker wanted to evoke. The “meaning”, therefore, of an utterance – and of the parts of an utterance, such as individual sentences and their parts, the individual words, etc. – is, in this view, some mental item. Successful communication requires that there be widespread agreement amongst speakers of the same public language as to the mental entities that are correlated with any particular public words.

Default reasoning occurs when the available information does not deductively guarantee the truth of the conclusion; and the conclusion is nonetheless correctly arrived at. The formalisms that have been developed in Artificial Intelligence to capture this mode of reasoning have suffered from a lack of agreement as to which non-monotonic inferences should be considered correct; and so Lifschitz 1989 produced a set of “Nonmonotonic Benchmark Problems” which all future formalisms are supposed to honor. The present work investigates the extent to which humans follow the prescriptions set out in these Benchmark Problems.

Average-NPs, such as the one in the title of this paper, have been claimed to be ‘linguistically identical’ to any other definite-NPs but at the same time to be ‘semantically inconsistent’ with these other definite-NPs. To some this is an ironclad proof of the irrelevance of semantics to linguistics. We argue that both of the initial claims are wrong: average-NPs are not ‘linguistically identical’ to other definite-NPs but instead show a number of interesting divergences, and we provide a plausible semantic account for them that is not ‘semantically inconsistent’ with the account afforded other definite-NPs but in fact blends quite nicely with one standard account of the semantics for NPs.

In this study, we examine the problem of belief revision, defined as deciding whic h of several initially-accepted sentences to disbelieve, when new information presents a l ogical inconsistency with the initial set. In the first three experiments, the initial sentence set included a conditional sentence, a non-conditional sentence, and an inferred conclusi on drawn from the first two. The new information contradicted the inferred conclusion. Results indicated that the conditional sentences were more readily abandoned than non-c onditional sentences, even when either choice would lead to a consistent belief state, and that this preference was more pronounced when problems used natural language cover sto ries rather than symbols. The pattern of belief revision choices differed depending on whe ther the contradicted conclusion from the initial belief set had been a modus ponens or m odus tollens inference. Two additional experiments examined alternative model-theoretic definitions of minimal change to a belief state, using problems that contained multiple mo dels of the initial belief state and of the new information that provided the contradiction. The results indicated that people did not follow any of four formal definitions of minimal change on these problems. The new information and the contradiction it offered was not, for example, used to select a particular model of the initial belief state as a way of reconci ling the contradiction. The preferred revision was to retain only those initial sentences th at had the same, unambiguous truth value within and across both the initial and new info rmation sets. The study and results are presented in the context of certain logic-based for malizations of belief revision, syntactic and model-theoretic representations of belief stat es, and performance models of human deduction. Principles by which some types of sent ences might be more "entrenched" than others in the face of contradiction are also discuss ed from the perspective of induction and theory revision.

Psychologism in logic is the doctrine that the semantic content of logical terms is in some way a feature of human psychology. We consider the historically influential version of the doctrine, Psychological Individualism, and the many counter-arguments to it. We then propose and assess various modifications to the doctrine that might allow it to avoid the classical objections. We call these Psychological Descriptivism, Teleological Cognitive Architecture, and Ideal Cognizers. These characterizations give some order to the wide range of modern views that are seen as psychologistic because of one or another feature. Although these can avoid some of the classic objections to psychologism, some still hold.

We investigate the notion of relevance as it pertains to ‘commonsense’, subjunctive conditionals. Relevance is taken here as a relation between a property (such as having a broken wing) and a conditional (such as birds typically fly). Specifically, we explore a notion of ‘causative’ relevance, distinct from ‘evidential’ relevance found, for example, in probabilistic approaches. A series of postulates characterising a minimal, parsimonious concept of relevance is developed. Along the way we argue that no purely logical account of relevance (even at the metalevel) is possible. Finally, and with minimal restrictions, an explicit definition that agrees with the postulates is given.

Advances in the underlying theory of a subdiscipline of AI can result in an apparently impressive improvement in the performance of a system that incorporates the advance. This impression typically comes from observing improved performance of the new system on some test problems. However, the improvement in performance may be for only the problems used in the testing, and performance on other problems may be degraded, possibly resulting overall in an degradation of the system’s performance. This comes about typically when the incorporation of the new feature increases the resources required overall, but the feature has bene- fits on only some problems (e.g., those used to test the new new system). In general, a localized theoretical advance is only rarely sufficient to increase the overall performance of any complex system. Therefore, researchers who make theoretical advances, also need some way to demonstrate that an advance really does have general, overall positive consequences for system..

Although advances in the underlying theory of a subdiscipline of AI can result in impressive increases in the performance of systems that employ such an underlying theory, this sometimes seems to be almost "by accident." The reason for this impression is that the impressive new advance in performance sometimes seems to be a feature merely of the specific example tests that are being demonstrated. And this impression is further strengthened when one notes that, in general, a localized theoretical advance is only rarely suflicient to increase the overall performance of any complex system. As a result of all these considerations, researchers who make theoretical advances are left desiring some method that would demonstrate that the advance really does have general, overall positive consequences for their system's performance.

This is a position paper concerning the role of empirical studies of human default reasoning in the formalization of AI theories of default reasoning. We note that AI motivates its theoretical enterprise by reference to human skill at default reasoning, but that the actual research does not make any use of this sort of information and instead relies on intuitions of individual investigators. We discuss two reasons theorists might not consider human performance relevant to formalizing default reasoning: (a) that intuitions are sufficient to describe a model, and (b) that human performance in this arena is irrelevant to a competence model of the phenomenon. We provide arguments against both these reasons. We then bring forward three further considerations against the use of intuitions in this arena: (a) it leads to an unawareness of predicate ambiguity, (b) it presumes an understanding of ordinary language statements of typicality, and (c) it is similar to discredited views in other fields. We advocate empirical investigation of the range of human phenomena that intuitively embody default reasoning. Gathering such information would provide data with which to generate formal default theories and against which to test the claims of proposed theories. Our position is that such data are the very phenomena that default theories are supposed to explain.

Agent communication languages (ACLs) invoke speech act theory and define individual message types by reference to particular combinations of beliefs and desires of the speaker (feasibility preconditions). Even when the mental states are restricted to a small set of nested beliefs, it seems that there might be a very large number of different possible preconditions, and therefore a very large number of different message types. With some constraints on the mental attitude of the speaker, we enumerate the possible belief states that could serve as preconditions for individual message types, and we identify how these states correspond to different possible message types. We then compare these with FIPA’s primitive message types. Our approach clarifies the nature of core message types in an ACL, and perhaps settles issues concerning just how many, and what types of, speech acts should be seen as primitive in such languages.

Computational semantics is the study of how to represent meaning in a way that computers can use. For the authors of this textbook, this study includes the representation of the meaning of natural language in logic formalisms, the recognition of certain relations that hold within this formalization (such as synonymy, consistency, and implication), and the computational implementation of all this. I think that, while there probably are not many courses devoted to computational semantics, this book could profitably be incorporated into more traditional computational linguistics courses, especially when two courses are offered serially. The material here could be spread out and integrated into parts of a more standard pair of these courses, and it would result in a substantial widening of the knowledge that students come away with from these courses.